Does Twitter language reliably predict heart disease? A commentary on Eichstaedt et al. (2015a)

Area of Science:

• Public Health
• Computational Social Science
• Epidemiology

Background:

• Eichstaedt et al. (2015a) claimed Twitter language predicts atherosclerotic heart disease (AHD) mortality at the US county level.
• The study suggested negative language correlated with higher AHD mortality and positive language with lower rates.

Purpose of the Study:

• To critically evaluate the methodology and findings of Eichstaedt et al. (2015a).
• To assess the robustness and generalizability of using aggregated Twitter data for predicting geographical variations in mortality.

Main Methods:

• Reproduced regression and correlation models from Eichstaedt et al. using their provided data.
• Substituted atherosclerotic heart disease (AHD) mortality with suicide mortality as the outcome variable.
• Examined assumptions regarding AHD, secondary data analysis, and the definition of communities at the county level.

Main Results:

• Replication of Eichstaedt et al.'s models showed reversed associations between language sentiment and mortality when suicide was the outcome variable.
• Identified significant conceptual and methodological limitations in the original study's approach.
• The predictive power of aggregated Twitter data for geographical AHD mortality variation was found to be unsubstantiated.

Conclusions:

• The methodology used by Eichstaedt et al. is not robust or generalizable for understanding geographical variations in AHD mortality.
• Bulk analysis of Twitter data, as performed, does not offer valuable insights into AHD mortality rates.
• Further research is needed to explore reliable methods for utilizing online data in public health research.